Fix collect() failure when pandas columns contain NULL values, date type mapping and tune-grid pipe placeholder compatibility

[tobiasdut]

Summary

This PR fixes three issues in Spark → Pandas → R collection.

---

1) NULL columns broke tibble conversion

Before

Collecting tables with NULL-heavy columns failed:

sparklyr::sdf_sql(sc, "SELECT * FROM "...") %>% head()

Error:

Error in `dplyr::as_tibble()`:
! All columns in a tibble must be vectors.
✖ Column `X` is NULL.
✖ Column `Y` is NULL.
...

Fix

to_pandas_cleaned() now flattens fallback output into proper R vectors while preserving row count, preventing literal NULL columns.

After

head() collects successfully without tibble errors.

---

2) Spark date returned as <dbl>

Before

Date columns sometimes arrived as numeric (days since epoch):

sparklyr::sdf_sql(sc, "...") %>% select(modification_date)

modification_date
<dbl>
17680
17301
...

Fix

If py_type == "date", values are explicitly converted via:

as.Date(x, origin = "1970-01-01")

across degraded R-side types (numeric, integer, character, list, logical).

After

modification_date
<date>
2018-05-29
2017-05-15
...

---

3) tune-grid.R pipe placeholder parse error

Before

... |> _$val

Could fail with:

pipe placeholder can only be used as a named argument

Fix

Replaced with parser-safe equivalent:

... |> (\(x) x$val)()

Behavior unchanged; compatibility improved.

[tobiasdut]

Related to #177

[dabruehl]

@edgararuiz Could you verify the fix and merge it please?

Thx @tobiasdut